In [275]:
import warnings
warnings.filterwarnings('ignore')
In [2]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
In [3]:
# Power_Actual DataSet
power = pd.read_csv("/Users/apple/Downloads/Assignment for Data Scientist/power_actual_v - power_actual.csv")
power.head()
Out[3]:
In [4]:
#shape of power
power.shape
Out[4]:
In [5]:
#Summary of Power DataSet
power.describe()
Out[5]:
Observed that "Power" variable is having some of the data points are in watt and most of them are Kilowatt. So, I have converted all the values into Kilowatt by using Excel.
In [6]:
#Removing unwanted Data Variables
power1=power.drop(["Unnamed: 0","ghi","gti","power"],axis=1)
power1.head()
Out[6]:
In [7]:
#datetime conversion
power1['datetime'] = pd.to_datetime(power1.datetime)
power1.head()
Out[7]:
In [8]:
power1['hours'] = power1['datetime'].dt.hour
power1.head()
Out[8]:
In [9]:
power1['date'] = power1['datetime'].dt.date
power1.head()
Out[9]:
In [10]:
power1.set_index('datetime',inplace=True)
In [11]:
power1['power1'].plot(figsize=(25,10)).plot(label='Power for every 15 mins')
plt.title('Power Generated From OCT 2017 - SEPT 2019 for every 15 mins')
plt.xlabel("Date Time")
plt.ylabel("Power Generated watts")
plt.legend()
Out[11]:
In [12]:
power1['power1'].rolling(window=96).mean().plot(figsize=(25,10))
plt.title('rolling mean of ( 96 = 24 hours ) Power Generated From OCT 2017 - SEPT 2019 for every 15 mins')
plt.xlabel("Date Time")
plt.ylabel("Power Generated watts")
plt.legend()
Out[12]:
In [13]:
power1['date'].nunique()
Out[13]:
In [14]:
power1.dtypes
Out[14]:
In [15]:
power1.info()
In [16]:
power1.isnull().sum()
Out[16]:
In [17]:
# Finding number of null values by Plot
import missingno
missingno.matrix(power1, figsize = (10,3))
Out[17]:
In [18]:
# Power if sumed up for each day by using groupby . so that we have one value for each day..reset_index()
power2=power1.groupby([power1['date']])['power1'].sum().to_frame()
power2
Out[18]:
In [ ]:
In [19]:
power2.shape
Out[19]:
In [20]:
power2['power1'].nunique()
Out[20]:
In [21]:
power2['power1'].describe()
Out[21]:
In [22]:
#Histogam for "power"
power2["power1"].hist(figsize = (15,5), color='#F39C12')
#plt.label("power ")
plt.title("Histogram Power per day")
plt.xlabel("Power")
plt.ylabel("Count")
plt.show()
In [23]:
power2.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('Power Generated From OCT 2017 - SEPT 2019 for each day ')
plt.ylabel('Power per day (KW)')
Out[23]:
In [24]:
# Rolling mean of 10 days
power_mean =power2['power1'].rolling(window=10).mean()
power_mean.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('Rolling mean of 10 days Power Generated From OCT 2017 - SEPT 2019 for each day')
plt.ylabel('Power per day (KW)')
Out[24]:
Weather Actual¶
In [25]:
weather = pd.read_csv("/Users/apple/Downloads/Assignment for Data Scientist/weather_actuals - weather_actuals.csv")
weather.head()
Out[25]:
Replaced "-9999" with NaN value with the help of Google Spread Sheets.
In [26]:
weather.columns
Out[26]:
In [27]:
weather.shape
Out[27]:
In [28]:
weather.isnull().sum()
Out[28]:
In [29]:
# Finding null values by Plot
missingno.matrix(weather, figsize = (15,3) )
Out[29]:
In [30]:
weather.describe()
Out[30]:
In [31]:
weather['wind_gust'].describe()
Out[31]:
In [32]:
weather['heat_index'].describe()
Out[32]:
In [33]:
weather['pressure'].describe()
Out[33]:
In [34]:
weather1=weather.drop(["Unnamed: 0",'plant_id','wind_chill','heat_index','qpf','snow', 'pop', 'fctcode','precip_accumulation'],axis=1)
weather1.head()
Out[34]:
In [35]:
weather1.columns
Out[35]:
In [36]:
weather1.shape
Out[36]:
In [37]:
weather1.info()
In [38]:
weather1.isnull().sum()
Out[38]:
In [39]:
# Finding number of null values by Plot
missingno.matrix(weather1, figsize = (10,3))
Out[39]:
In [40]:
from sklearn.impute import SimpleImputer
from numpy import nan
In [41]:
weather1['cloud_cover'] = weather1['cloud_cover'].fillna(0.23)
In [42]:
weather1['cloud_cover'].isnull().sum()
Out[42]:
In [43]:
weather1['wind_bearing'] = weather1['wind_bearing'].fillna(188)
In [44]:
weather1['wind_bearing'].isnull().sum()
Out[44]:
In [45]:
weather1['wind_speed'] = weather1['wind_speed'].fillna(2.63)
In [46]:
weather1['wind_speed'].isnull().sum()
Out[46]:
In [47]:
weather1['wind_gust'] = weather1['wind_gust'].fillna(4.24)
In [48]:
weather1['wind_gust'].isnull().sum()
Out[48]:
In [49]:
weather1['pressure'] = weather1['pressure'].fillna(1009.07)
In [50]:
weather1['pressure'].isnull().sum()
Out[50]:
In [51]:
weather1['uv_index'] = weather1['uv_index'].fillna(1)
In [52]:
weather1['uv_index'].isnull().sum()
Out[52]:
In [53]:
weather1['ozone'] = weather1['ozone'].fillna(274.4)
In [54]:
weather1['ozone'].isnull().sum()
Out[54]:
In [55]:
weather1['precip_intensity'] = weather1['precip_intensity'].fillna(0.05)
In [56]:
weather1['precip_intensity'].isnull().sum()
Out[56]:
In [57]:
weather1['precip_probability'] = weather1['precip_probability'].fillna(0.01)
In [58]:
weather1['precip_probability'].isnull().sum()
Out[58]:
In [59]:
weather1['precip_type'].unique()
Out[59]:
In [60]:
weather1['precip_type'] = weather1['precip_type'].fillna('no rain')
In [61]:
weather1['precip_type'].isnull().sum()
Out[61]:
In [62]:
weather1['visibility'] = weather1['visibility'].fillna(12.2)
In [63]:
weather1['visibility'].isnull().sum()
Out[63]:
In [64]:
weather1.isnull().sum()
Out[64]:
In [65]:
# Finding number of null values by Plot
missingno.matrix(weather1, figsize = (10,3))
Out[65]:
In [66]:
weather1.describe()
Out[66]:
In [67]:
weather1.columns
Out[67]:
In [68]:
#datetime conversion
weather1['datetime_utc'] = pd.to_datetime(weather1.datetime_utc)
weather1['datetime_local'] = pd.to_datetime(weather1.datetime_local)
weather1['sunrise'] = pd.to_datetime(weather1.sunrise)
weather1['sunset'] = pd.to_datetime(weather1.sunset)
weather1['updated_at'] = pd.to_datetime(weather1.updated_at)
In [69]:
weather1.head()
Out[69]:
In [70]:
Act_weather=weather1.copy()
In [71]:
Act_weather.head()
Out[71]:
In [72]:
Act_weather.shape
Out[72]:
In [73]:
Act_weather=Act_weather.drop(["datetime_utc","sunrise","sunset","updated_at"],axis=1)
Act_weather.head()
Out[73]:
In [74]:
Act_weather['date'] = Act_weather['datetime_local'].dt.date
Act_weather.head()
Out[74]:
In [75]:
Act_weather['datetime_local'].nunique()
Out[75]:
In [76]:
Act_weather['date'].nunique()
Out[76]:
In [77]:
Act_weather.columns
Out[77]:
In [78]:
Act_weather_df = Act_weather.groupby(['date'])[['cloud_cover', 'apparent_temperature', 'temperature',
'humidity', 'dew_point', 'wind_bearing', 'wind_speed', 'wind_gust',
'pressure', 'uv_index', 'ozone', 'precip_intensity',
'precip_probability', 'visibility']].mean()
In [79]:
Act_weather_df.head()
Out[79]:
In [80]:
Act_weather_df.to_csv('Act_weather_df.csv')
In [81]:
Act_weather_df.shape
Out[81]:
In [82]:
Act_weather_df1 = Act_weather.groupby(['date'])[['precip_type', 'icon', 'summary']].max()
In [83]:
Act_weather_df1
Out[83]:
In [84]:
Actual_weather=Act_weather_df.join(Act_weather_df1)
Actual_weather.head()
Out[84]:
In [85]:
# joining Actual_weather and Actual_Power Datasets
dataframe=Actual_weather.join(power2)
dataframe.head()
Out[85]:
In [86]:
dataframe.shape
Out[86]:
In [87]:
from sklearn.preprocessing import LabelEncoder
labelencoder=LabelEncoder()
dataframe['icon']=labelencoder.fit_transform(dataframe['icon'])
dataframe['summary']=labelencoder.fit_transform(dataframe['summary'])
In [88]:
dataframe=pd.get_dummies(dataframe,columns=["precip_type"],drop_first=True)
dataframe.head()
Out[88]:
In [89]:
dataframe.columns
Out[89]:
In [90]:
dataframe.dtypes
Out[90]:
In [91]:
# Finding number of null values by Plot
missingno.matrix(dataframe, figsize = (10,3))
Out[91]:
pairplot¶
In [92]:
#pairplot
sns.pairplot(dataframe)
In [93]:
dataframe['cloud_cover'].plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('cloud_cover From OCT 2017 - SEPT 2019 ')
plt.ylabel('Cloud unit measure ( okta )')
Out[93]:
In [94]:
cloud_mean =dataframe['cloud_cover'].rolling(window=10).mean()
cloud_mean.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('cloud_cover From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('Cloud unit measure ( okta )')
Out[94]:
In [95]:
apparent_temperature =dataframe['apparent_temperature'].rolling(window=30).mean()
temperature =dataframe['temperature'].rolling(window=10).mean()
apparent_temperature.plot(grid=True, figsize=(25,10), legend=True, color='#2980B9')
temperature.plot(grid=True, figsize=(25,10), legend=True, color='#E74C3C')
plt.title('apparent_temperature vs Temperature From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('Temperature (Celcius)')
Out[95]:
In [96]:
humidity =dataframe['humidity'].rolling(window=10).mean()
humidity.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('humidity From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('humidity value')
Out[96]:
In [97]:
dew_point =dataframe['dew_point'].rolling(window=10).mean()
dew_point.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('dew_point From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('dew_point (celcius)')
Out[97]:
In [98]:
wind_speed =dataframe['wind_speed'].rolling(window=10).mean()
wind_speed.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('wind_speed From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('wind_speed (Km/h)')
Out[98]:
In [99]:
wind_gust =dataframe['wind_gust'].rolling(window=10).mean()
wind_gust.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('wind_gust From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('wind_gust (m/s)')
Out[99]:
In [100]:
pressure =dataframe['pressure'].rolling(window=10).mean()
pressure.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('pressure From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('pressure (Hectapascal (hPa))')
Out[100]:
In [101]:
uv_index =dataframe['uv_index'].rolling(window=10).mean()
uv_index.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('uv_index From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('uv_index ')
Out[101]:
In [102]:
ozone =dataframe['ozone'].rolling(window=10).mean()
ozone.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('ozone From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('ozone value ')
Out[102]:
In [103]:
precip_intensity =dataframe['precip_intensity'].rolling(window=10).mean()
precip_intensity.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('precip_intensity From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('precip_intensity (mm/h) ')
Out[103]:
In [104]:
precip_probability =dataframe['precip_probability'].rolling(window=10).mean()
precip_probability.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('precip_probability From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('precip_probability (%) ')
Out[104]:
In [105]:
visibility =dataframe['visibility'].rolling(window=10).mean()
visibility.plot(grid=True, figsize=(25,10), legend=True, color='#EC7063')
plt.title('visibility From OCT 2017 - SEPT 2019 for every 10 days ')
plt.ylabel('visibility ')
Out[105]:
In [106]:
# Correlation with power
dataframe.corr()['power1']
Out[106]:
In [107]:
import seaborn as sns
corr=dataframe.iloc[:,:-1].corr()
top_features=corr.index
plt.figure(figsize=(20,20))
sns.heatmap(dataframe[top_features].corr(),annot=True)
Out[107]:
Feature Selection¶
In [108]:
X=dataframe.drop(['power1'],axis=1)
y=dataframe['power1']
In [109]:
from sklearn.feature_selection import mutual_info_regression
mutual_info=mutual_info_regression(X,y)
In [110]:
mutual_data=pd.Series(mutual_info,index=X.columns)
mutual_data.sort_values(ascending=False)
Out[110]:
In [111]:
threshold=0.8
In [112]:
# find and remove correlated features
def correlation(dataset, threshold):
col_corr = set() # Set of all the names of correlated columns
corr_matrix = dataset.corr()
for i in range(len(corr_matrix.columns)):
for j in range(i):
if abs(corr_matrix.iloc[i, j]) > threshold: # we are interested in absolute coeff value
colname = corr_matrix.columns[i] # getting the name of column
col_corr.add(colname)
return col_corr
In [113]:
correlation(X,threshold)
Out[113]:
ExtraTreesRegressor¶
In [114]:
from sklearn.ensemble import ExtraTreesRegressor
import matplotlib.pyplot as plt
model=ExtraTreesRegressor()
model.fit(X,y)
Out[114]:
In [115]:
print(model.feature_importances_)
In [116]:
ranked_features=pd.Series(model.feature_importances_,index=X.columns)
ranked_features.nlargest(10).plot(kind='barh')
plt.show()
In [117]:
dataframe.columns
Out[117]:
In [118]:
final_weather=pd.DataFrame(dataframe,columns=['humidity','uv_index', 'ozone','temperature','visibility', 'wind_bearing', 'pressure','apparent_temperature','wind_gust','precip_probability','power1'])
final_weather.head()
Out[118]:
Weather Forecasted¶
In [119]:
wf=pd.read_csv("/Users/apple/Downloads/Assignment for Data Scientist/weather_forecast.csv")
wf.head()
Out[119]:
In [120]:
wf.shape
Out[120]:
In [121]:
wf.isnull().sum()
Out[121]:
In [122]:
# Finding number of null values by Plot
missingno.matrix(wf, figsize = (10,3))
Out[122]:
In [123]:
wf=wf.drop(["Unnamed: 0",'plant_id','wind_chill','heat_index','qpf','snow', 'pop', 'fctcode','precip_accumulation'],axis=1)
wf.head()
Out[123]:
In [124]:
# Finding number of null values by Plot
missingno.matrix(wf, figsize = (10,3))
Out[124]:
In [125]:
wf['precip_type'].unique()
Out[125]:
In [126]:
wf['precip_type'] = wf['precip_type'].fillna('no rain')
In [127]:
wf.columns
Out[127]:
In [128]:
wf['datetime_utc'] = pd.to_datetime(wf.datetime_utc)
wf['datetime_local'] = pd.to_datetime(wf.datetime_local)
wf['sunrise'] = pd.to_datetime(wf.sunrise)
wf['sunset'] = pd.to_datetime(wf.sunset)
wf['updated_at'] = pd.to_datetime(wf.updated_at)
In [129]:
wf.head()
Out[129]:
In [130]:
wf['date']=wf['datetime_utc'].dt.date
In [131]:
For_date=wf['date']
For_date
Out[131]:
In [132]:
For_weather=wf.copy()
In [133]:
For_weather=For_weather.drop(["datetime_utc","sunrise","sunset","updated_at"],axis=1)
For_weather.head()
Out[133]:
In [134]:
For_weather_df = For_weather.groupby(['date'])['cloud_cover', 'apparent_temperature', 'temperature',
'humidity', 'dew_point', 'wind_bearing', 'wind_speed', 'wind_gust',
'pressure', 'uv_index', 'ozone', 'precip_intensity',
'precip_probability', 'visibility'].mean().reset_index()
For_weather_df.head()
Out[134]:
In [135]:
forecast_date=For_weather_df.iloc[:,0:1]
forecast_date.head()
Out[135]:
In [136]:
For_weather_df.set_index('date',inplace=True)
For_weather_df.head()
Out[136]:
In [137]:
For_weather_df1 = For_weather.groupby(['date'])[['precip_type', 'icon', 'summary']].max()
In [138]:
Forecast_weather=For_weather_df.join(For_weather_df1)
Forecast_weather.head()
Out[138]:
In [139]:
Forecast_weather['icon']=labelencoder.fit_transform(Forecast_weather['icon'])
Forecast_weather['summary']=labelencoder.fit_transform(Forecast_weather['summary'])
In [140]:
Forecast_weather=pd.get_dummies(Forecast_weather,columns=["precip_type"],drop_first=True)
Forecast_weather.head()
Out[140]:
In [141]:
# Finding number of null values by Plot
missingno.matrix(Forecast_weather, figsize = (10,3))
Out[141]:
Pairplot¶
In [142]:
# Pairplot
sns.pairplot(Forecast_weather)
In [143]:
ff_weather=pd.DataFrame(Forecast_weather,columns=['humidity','uv_index', 'ozone','temperature','visibility', 'wind_bearing', 'pressure','apparent_temperature','wind_gust','precip_probability'])
ff_weather.head()
Out[143]:
In [145]:
ff_weather.shape
Out[145]:
Spliting Data¶
In [146]:
x_train=final_weather.iloc[:,:-1]
y_train=final_weather.iloc[:,-1:]
In [147]:
x_train.head()
Out[147]:
In [148]:
y_train.head()
Out[148]:
Feature Preprocessing¶
In [149]:
from sklearn import preprocessing
In [170]:
n_x_train=preprocessing.normalize(x_train)
n_x_train
Out[170]:
In [176]:
n_ff_weather=preprocessing.normalize(ff_weather)
n_ff_weather
Out[176]:
In [171]:
from sklearn.decomposition import PCA
pca = PCA(n_components=3)
lr_pca=pca.fit(n_x_train)
lr_pca
Out[171]:
Linear Regression Model¶
In [174]:
from sklearn.linear_model import LinearRegression
regression=LinearRegression()
regression.fit(n_x_train,y_train)
Out[174]:
In [178]:
y_pred_lr=regression.predict(n_ff_weather)
y_pred_lr
Out[178]:
In [179]:
score_lr=regression.score(n_ff_weather,y_pred_lr)
score_lr
Out[179]:
In [188]:
#from sklearn.metrics import mean_squared_error
#mean_squared_error(y_train,y_pred_lr)
In [182]:
lr_Power=pd.DataFrame(y_pred_lr,columns=['LRF_Power'])
lr_Power.head()
Out[182]:
In [183]:
# joining forecast_date and lr_Power Datasets
lr_fp=forecast_date.join(lr_Power)
lr_fp.head()
Out[183]:
In [295]:
lr_fp.to_csv('Linear_Regression_Result.csv')
In [184]:
lr_fp.set_index('date',inplace=True)
lr_fp.head()
Out[184]:
In [185]:
# Power Generated from 1st oct 2019 - 27th oct 2019
final_weather['power1'].plot(grid=True, figsize=(25,10), label='Power (oct 2017 - sept 2019)', color='#2980B9')
lr_fp['LRF_Power'].plot(grid=True, figsize=(25,10), label='Forecasted Power (1st oct 2019 - 27th oct 2019) ', color='#E74C3C')
plt.title('Forecasted Linear Regression Power Generated from 1st oct 2019 - 27th oct 2019')
plt.xlabel("Date")
plt.ylabel("Power (Kw/h)")
plt.legend()
Out[185]:
Random Forest¶
In [197]:
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import RandomForestRegressor
In [208]:
# Create a pipeline
pipe = Pipeline([("regressor", RandomForestRegressor())])
# hyperparameters
grid_param = [
{"regressor": [RandomForestRegressor()],
"regressor__n_estimators": [10, 100, 1000],
"regressor__max_depth":[5,8,15,25,30,None],
"regressor__min_samples_leaf":[1,2,5,10,15,100],
"regressor__max_leaf_nodes": [2, 5,10]}]
RF_model = GridSearchCV(pipe, grid_param, cv=5, verbose=0,n_jobs=-1) # Fit grid search
RF_model.fit(x_train,y_train)
Out[208]:
In [209]:
print(RF_model.best_estimator_)
In [215]:
y_pred_RF=RF_model.predict(ff_weather)
y_pred_RF
Out[215]:
In [216]:
score_RF=RF_model.score(ff_weather,y_pred_RF)
score_RF
Out[216]:
In [218]:
RF_FPower=pd.DataFrame(y_pred_RF,columns=['RFF_Power'])
RF_FPower.head()
Out[218]:
In [219]:
# joining forecast_date and RF_FPower Datasets
RF_fp=forecast_date.join(RF_FPower)
RF_fp.head()
Out[219]:
In [296]:
RF_fp.to_csv('RandomForest_Result.csv')
In [220]:
RF_fp.set_index('date',inplace=True)
RF_fp.head()
Out[220]:
In [221]:
# Power Generated from 1st oct 2019 - 27th oct 2019
final_weather['power1'].plot(grid=True, figsize=(25,10), label='Power (oct 2017 - sept 2019)', color='#2980B9')
RF_fp['RFF_Power'].plot(grid=True, figsize=(25,10), label='Forecasted Power (1st oct 2019 - 27th oct 2019) ', color='#E74C3C')
plt.title('Random Forest Power Generated from 1st oct 2019 - 27th oct 2019')
plt.xlabel("Date")
plt.ylabel("Power (Kw/h)")
plt.legend()
Out[221]:
XGBoost Model¶
In [222]:
import xgboost as xgb
In [224]:
# Create a pipeline
pipe = Pipeline([("regressor", xgb.XGBRegressor())])
# Create dictionary with candidate learning algorithms and their hyperparameters
grid_param = [
{"regressor": [xgb.XGBRegressor()],
"regressor__n_estimators": [10, 50, 100],
"regressor__max_depth":[5,8,15,25,30,None],
"regressor__learning_rate":[0,0.1,0.2,0.3,0.5,0.8,1],
}]
# create a gridsearch of the pipeline, the fit the best model
XGB_model = GridSearchCV(pipe, grid_param, cv=5, verbose=0,n_jobs=-1) # Fit grid search
XGB_model.fit(x_train,y_train)
Out[224]:
In [225]:
print(XGB_model.best_estimator_)
In [226]:
y_pred_XGB=XGB_model.predict(ff_weather)
y_pred_XGB
Out[226]:
In [227]:
score_XGB=XGB_model.score(ff_weather,y_pred_XGB)
score_XGB
Out[227]:
In [234]:
#rmse = np.sqrt(mean_squared_error(y_train,y_pred_XGB))
In [228]:
XGB_FPower=pd.DataFrame(y_pred_XGB,columns=['XGB_Power'])
XGB_FPower.head()
Out[228]:
In [229]:
# joining forecast_date and XGB_FPower Datasets
XGB_fp=forecast_date.join(XGB_FPower)
XGB_fp.head()
Out[229]:
In [298]:
XGB_fp.to_csv('XGBoost1_Result.csv')
In [230]:
XGB_fp.set_index('date',inplace=True)
XGB_fp.head()
Out[230]:
In [232]:
# Power Generated from 1st oct 2019 - 27th oct 2019
final_weather['power1'].plot(grid=True, figsize=(25,10), label='Power (oct 2017 - sept 2019)', color='#2980B9')
XGB_fp['XGB_Power'].plot(grid=True, figsize=(25,10), label='Forecasted Power (1st oct 2019 - 27th oct 2019) ', color='#E74C3C')
plt.title('XGBoost Power Generated from 1st oct 2019 - 27th oct 2019')
plt.xlabel("Date")
plt.ylabel("Power (Kw/h)")
plt.legend()
Out[232]:
SVM Model¶
In [236]:
from sklearn.svm import SVR
In [237]:
# Create a pipeline
svm_pipe = Pipeline([("regressor", SVR())])
# Create hyperparameters
grid_param = [
{"regressor": [SVR()],
"regressor__kernel": ['linear', 'poly','rbf'],
"regressor__gamma": [1e-4, 1e-3, 0.01, 0.1, 0.2, 0.5, 0.6, 0.9],
"regressor__C": [1, 10, 100, 1000, 10000],
}]
SVM_model = GridSearchCV(svm_pipe, grid_param, cv=5, verbose=0,n_jobs=-1) # Fit grid search
SVM_model.fit(n_x_train,y_train)
Out[237]:
In [238]:
print(SVM_model.best_estimator_)
In [239]:
y_pred_SVM=SVM_model.predict(n_ff_weather)
y_pred_SVM
Out[239]:
In [240]:
score_SVM=SVM_model.score(n_ff_weather,y_pred_SVM)
score_SVM
Out[240]:
In [241]:
SVM_FPower=pd.DataFrame(y_pred_SVM,columns=['SVM_Power'])
SVM_FPower.head()
Out[241]:
In [243]:
# joining forecast_date and SVM_Forecasted Power Datasets
SVM_fp=forecast_date.join(SVM_FPower)
SVM_fp.head()
Out[243]:
In [299]:
SVM_fp.to_csv('SVR_Result.csv')
In [244]:
SVM_fp.set_index('date',inplace=True)
SVM_fp.head()
Out[244]:
In [245]:
# Power Generated from 1st oct 2019 - 27th oct 2019
final_weather['power1'].plot(grid=True, figsize=(25,10), label='Power (oct 2017 - sept 2019)', color='#2980B9')
SVM_fp['SVM_Power'].plot(grid=True, figsize=(25,10), label='Forecasted Power (1st oct 2019 - 27th oct 2019) ', color='#E74C3C')
plt.title('SVM Power Generated from 1st oct 2019 - 27th oct 2019')
plt.xlabel("Date")
plt.ylabel("Power (Kw/h)")
plt.legend()
Out[245]:
SARIMA¶
ARMA - Autoregression (P) Integrated (d) Moving Average(q)
In [246]:
#plot_acf & plot_pacf
from statsmodels.graphics.tsaplots import plot_acf,plot_pacf
In [247]:
#plot_acf - (q)
#plot_pacf - (p)
plot_acf(final_weather['power1'])
plot_pacf(final_weather['power1'])
Out[247]:
p = 3 ; d=1 ; q = 9 ; [ d=0(no stationarity ) d=1(stationarity ) ]
In [252]:
#Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors model [SARIMAX]
import statsmodels.api as sm
SARIMA_model=sm.tsa.statespace.SARIMAX(final_weather['power1'],order=(3, 1, 9),seasonal_order=(1,1,1,9))
results=SARIMA_model.fit()
In [284]:
SARIMA_FPower=results.predict(start=568,end=597,dynamic=True)
SARIMA_FPower
Out[284]:
In [300]:
SARIMA_FPower.to_csv('SARIMAA_Result.csv')
Neural Networks¶
In [285]:
from sklearn.neural_network import MLPRegressor
In [286]:
# Create a pipeline
MLP_pipe = Pipeline([("regressor", MLPRegressor())])
# Create hyperparameters
grid_param = [
{"regressor": [MLPRegressor()],
"regressor__activation": ['relu','identity', 'tanh','logistic'],
"regressor__solver": ['lbfgs','sgd', 'adam'],
"regressor__learning_rate": ['constant','invscaling','adaptive'],
"regressor__max_iter":[200,300,400,500,1000]
}]
MLP_model = GridSearchCV(MLP_pipe, grid_param, cv=5, verbose=0,n_jobs=-1) # Fit grid search
MLP_model.fit(n_x_train,y_train)
Out[286]:
In [287]:
print(MLP_model.best_estimator_)
In [288]:
y_pred_MLP=MLP_model.predict(n_ff_weather)
y_pred_MLP
Out[288]:
In [289]:
score_MLP=MLP_model.score(n_ff_weather,y_pred_MLP)
score_MLP
Out[289]:
In [290]:
MLP_FPower=pd.DataFrame(y_pred_MLP,columns=['MLP_Power'])
MLP_FPower.head()
Out[290]:
In [291]:
# joining forecast_date and MLP_FPower Power Datasets
MLP_fp=forecast_date.join(MLP_FPower)
MLP_fp.head()
Out[291]:
In [301]:
MLP_fp.to_csv('MLP_result.csv')
In [292]:
MLP_fp.set_index('date',inplace=True)
MLP_fp.head()
Out[292]:
In [293]:
# Power Generated from 1st oct 2019 - 27th oct 2019
final_weather['power1'].plot(grid=True, figsize=(25,10), label='Power (oct 2017 - sept 2019)', color='#2980B9')
MLP_fp['MLP_Power'].plot(grid=True, figsize=(25,10), label='Forecasted Power (1st oct 2019 - 27th oct 2019) ', color='#E74C3C')
plt.title('MLP Power Generated from 1st oct 2019 - 27th oct 2019')
plt.xlabel("Date")
plt.ylabel("Power (Kw/h)")
plt.legend()
Out[293]:
In [ ]: